Design method and system of an artificial intelligence agent based on user behavior

Abstract

A method for designing an AI agent based on user behavior includes: Step S1, configuring static user attributes for a specified AI agent; S2, configuring the operation process to simulate normal user behavior; S3, saving the AI ​​agent's behavior data during operation; S4, collecting trending comments from the platform and sending the comments to the specified platform according to specified parameters; S5, analyzing the impact of recommendation algorithms on user behavior based on the recorded data. This invention also includes a system for an AI agent based on user behavior, comprising: an attribute configuration module, a process customization module, a data storage module, a corpus generation module, and a behavior analysis module. This invention can simulate the behavior of normal users on various social media platforms and achieve fully automated operation. Through experimental data, the impact of recommendation algorithms on users can be analyzed, thereby revealing ways to avoid the negative impact of recommendation algorithms on users.

Description

Technical Field

[0001] This invention relates to the fields of data mining and machine learning technologies, and in particular to a design method and system for an artificial intelligence agent based on user behavior. Background Technology

[0002] With the development of big data, cloud computing, and deep learning technologies, public opinion dissemination has entered the 3.0 intelligent era, characterized by ubiquitous media, human-machine symbiosis, and algorithmic dominance. In this intelligent era, information dissemination and opinion interaction are largely controlled by media platform recommendation algorithms. However, modern artificial intelligence technology is typically a black box model, and the algorithms themselves are constantly evolving through interactions with hundreds of millions of users, thus forming an unprecedentedly complex human-machine coupled system with users and the media environment. Therefore, effective evaluation methods for the impact of recommendation algorithms are still lacking.

[0003] Patent CN201910585586.5 proposes a social engineering robot simulation method and equipment based on user attributes, but the method is only applicable to a single object and the robot has low flexibility. Summary of the Invention

[0004] This invention aims to overcome the aforementioned shortcomings of existing technologies, expand its application scope, and provide a design method and system for an artificial intelligence agent based on user behavior, effectively improving data security and providing an evaluation method for the impact on recommendation algorithms.

[0005] This invention combines human behavior on various social media platforms with probabilistic statistics and natural language processing methods to simulate human behavior and measure the impact of recommendation algorithms.

[0006] The technical solution adopted by the present invention to achieve the above-mentioned objectives is as follows:

[0007] The present invention provides a method for designing an AI agent based on user behavior, comprising the following steps:

[0008] S1: Configure user static attributes for a specified AI agent to form the agent's identity characteristics;

[0009] S2: Configure the operation process for the AI ​​agent to simulate normal user behavior;

[0010] S3: Save the behavioral data of the specified agent during its operation;

[0011] S4: Collect trending topics from the platform and send comments to the designated platform based on specified parameters;

[0012] S5: Analyze the impact of recommendation algorithms on user behavior based on the recorded data of the AI ​​agent;

[0013] Preferably, in step S1: setting static attributes of the AI ​​agent to depict the user's profile, including at least one of the following: name, gender, age, education, geographical location, hobbies and personal description, etc., to form the identity features of the AI ​​agent.

[0014] Preferably, in step S2: setting behavioral parameters for the AI ​​agent, including daily active time periods, daily number of actions, weekly active days, etc.; customizing the AI ​​agent's operation mode, including scheduling module, activity selection module, reading module, recommendation browsing module, collection browsing module, retrieval browsing module, attention browsing module, attention module, observation module, attack module, etc.; and customizing AI agents with different behaviors based on the combination of different modules.

[0015] Preferably, in step S3: the agent's behavioral data is recorded, including action type, information data, and comment content; wherein the action type includes at least one of the following: dormant, on, auto-follow / unfollow, information reading, comment, like, favorite, and forward; the information data includes at least one of the following: information title, text content, text length, text type, publication time, publisher, platform, number of views, number of comments, number of forwards, number of likes, number of favorites, and number of shares; the comment content includes at least one of the following: comment time, comment content, comment attitude, text, and commenting user.

[0016] Preferably, in step S4:

[0017] S4.1: On a specified platform, capture the content of popular users and popular comments under the topics you are interested in;

[0018] S4.2: Compile the collected text content into a dictionary, perform weight statistics and sorting, and generate a corpus;

[0019] S4.3: Based on the corresponding parameters, during the comment operation, determine the comment content according to a random function and send the comment to the target news or video on the specified platform.

[0020] Preferably, in step S5:

[0021] S5.1: Based on the behavioral data of the AI ​​agent, quantify the effectiveness of the target AI agent's behavior using modified conditional entropy and relative equality indicators;

[0022] S5.2: Based on the browsing records of the AI ​​agents, use text topic recognition algorithms, text sentiment recognition algorithms, and comment opinion tendency algorithms to analyze the impact of recommendation algorithms on different AI agent groups, thereby determining the behavioral trends of the AI ​​agents.

[0023] The present invention also provides a system for an artificial intelligence agent based on user behavior, comprising: an attribute configuration module, a process customization module, a data storage module, a corpus generation module, and a behavior analysis module;

[0024] The attribute configuration module configures user profiles for the AI ​​agent. Each natural person has many attributes, including name, gender, age, occupation, education, geographical location, hobbies and personal description. By inputting parameters, the user group that the AI ​​agent should imitate is determined, thereby improving the realism of the AI ​​agent.

[0025] The process customization module combines the behavioral processes of the AI ​​agent, mainly including a scheduling module, an activity selection module, a reading module, a browsing module, a following module, an observation module, and an attack module. The scheduling module sets the platform to which the AI ​​agent belongs, the activity time, and the waiting time. The activity selection module sets the browsing section, section weight, and repetition count. The browsing section sets the users to follow, the topics, and keywords. The following section sets the followed usernames. The reading section sets the reading time, and the weights for likes, favorites, comments, and shares. The AI ​​agent performs different actions according to different module combinations and with parameters configured in the corresponding modules.

[0026] The data storage module saves the user profile of the intelligent agent, the intelligent agent's operational behavior, the information data read, and the comment content into the database.

[0027] The corpus generation module captures user comments and popular comments under the topics of interest on the target platform, uses them as expected content, constructs a dictionary, sorts them according to weight, and selects the comments with the higher proportion as the comment content of the intelligent agent;

[0028] The behavior analysis module cleans the data in the database, uses text topic recognition algorithm, text sentiment recognition algorithm and comment opinion tendency algorithm to determine the impact of the platform's recommendation algorithm on the target intelligent agent, and visualizes it in charts;

[0029] The attribute configuration module, process customization module, data storage module, corpus generation module, and behavior analysis module are connected in sequence.

[0030] The working principle of this invention is:

[0031] By mimicking the real behavior of natural persons on social media platforms and using natural language processing and deep learning technologies, automated programs are written to enable artificial intelligence agents to operate on social media platforms.

[0032] Based on Chrome-based technology, the system can be integrated across different social media platforms, enabling click and touch operations through the positioning of web page elements. When commenting, a web crawler first collects the corpus, then natural language processing is used to generate appropriate comments for submission. The collected data is analyzed using deep learning technology to determine opinions and analyze behaviors.

[0033] The beneficial effects of this invention are:

[0034] (1) Based on user behavior, a customized intelligent agent is provided to fully and automatically simulate human behavior on social platforms, which can efficiently evade platform detection and provide a brand-new simulation experiment method.

[0035] (2). By analyzing agent behavior data, determine the impact of recommendation algorithms on users and predict the behavioral trends of agents.

[0036] (3) The system is easy to operate, and the operation process of the intelligent agent is set up by building blocks. Attached Figure Description

[0037] Figure 1 This is a flowchart of the method of the present invention;

[0038] Figure 2 This is a schematic diagram illustrating the simulation principle of the present invention;

[0039] Figure 3 This is the interface design diagram of the process customization module of the present invention;

[0040] Figure 4 This is a system structure block diagram of the present invention. Detailed Implementation

[0041] To better illustrate the purpose, technical process, and advantages of this invention, the following description, in conjunction with specific examples and accompanying drawings, further explains the invention. The illustrative examples and descriptions are for explaining the invention and are not intended to limit it.

[0042] To address the lack of effective evaluation methods for the impact of existing recommendation algorithms, this paper considers simulating user operations or behaviors on social platforms to analyze the impact of recommendation algorithms on human behavior.

[0043] To address this, a design method for AI agents based on user behavior is provided for experimental analysis.

[0044] like Figure 1 As shown, the design method for an AI agent based on user behavior includes the following steps:

[0045] S1: Configure static user attributes for a specified AI agent to create a user profile;

[0046] S2: Configure the operation process for the AI ​​agent to simulate normal user operations;

[0047] S3: Save the behavioral data of the specified agent during its operation;

[0048] S4: Collect trending comments from the platform and send the comments to the specified platform based on the specified parameters;

[0049] S5: Analyze the impact of recommendation algorithms on user behavior based on the recorded data of the AI ​​agent;

[0050] In step S1: Set the static attributes of the AI ​​agent to create a user profile, including at least one of the following: name, gender, age, education, geographical location, hobbies, and personal description.

[0051] In step S2: Set the behavioral parameters of the AI ​​agent, including daily active time periods, daily action frequency, and weekly active days; for example, a normal person's daily activity periods are 8:00-9:00 AM, 12:30-2:00 PM, and 7:00-12:00 PM, with an action frequency in the range of 10-50, and the average number of active days per week depends on the actual situation; customize the AI ​​agent's operation mode, including scheduling modules, activity selection modules, reading modules, recommendation browsing modules, collection browsing modules, retrieval browsing modules, attention browsing modules, attention modules, observation modules, attack modules, etc.; each module has different functions, and different AI agents with different behaviors are customized according to the combination of different modules.

[0052] In step S3: Record the agent's behavioral data, including action type, information data, and comment content; where action type includes at least one of the following: sleep, auto-follow / unfollow, information reading, comment, like, favorite, and forward; information data includes at least one of the following: information title, text content, text length, text type, publication time, publisher, platform, number of views, number of comments, number of forwards, number of likes, number of favorites, and number of shares; comment content includes at least one of the following: comment time, comment content, comment attitude, text, and commenting user; unify the data format and store it in the database.

[0053] Step S4 specifically includes:

[0054] To make the agent behave more like a human, it needs to send certain content when making comments. To make the content look more like it was sent by a real person, we added a content generation design. During operation, the agent crawls existing expected content, analyzes and matches it, and generates the text content to be sent.

[0055] S4.1: On the specified platform, crawl the popular user and popular comment content under the topic of interest; specifically, on the target platform, collect the corpus content that meets the conditions through web crawling, including the speeches and popular comments of popular users;

[0056] S4.2: Compile the collected text content into a dictionary, perform weight statistics and sorting, and generate a corpus;

[0057] S4.3: Based on the corresponding parameters, during the comment operation, determine the comment content according to a random function and send the comment to the target news or video on the specified platform.

[0058] Step S5 specifically includes:

[0059] S5.1: Based on the behavioral data of the AI ​​agent, quantify the effectiveness of the target AI agent's behavior using modified conditional entropy and relative equality indicators;

[0060] S5.2: Based on the browsing records of the AI ​​agents, use text topic recognition algorithms, text sentiment recognition algorithms, and comment opinion tendency algorithms to analyze the impact of recommendation algorithms on different AI agent groups, thereby determining the behavioral trends of the AI ​​agents.

[0061] like Figure 4 As shown, the system for implementing the user behavior-based artificial intelligence agent of the present invention includes: an attribute configuration module, a process customization module, a data storage module, a corpus generation module, and a behavior analysis module;

[0062] The attribute configuration module configures user profiles for the AI ​​agent. Each natural person has many attributes, including name, gender, age, occupation, education, geographical location, hobbies and personal description. By inputting parameters, the user group that the AI ​​agent should imitate is determined, thereby improving the realism and concealment of the AI ​​agent.

[0063] The process customization module combines the behavioral processes of the AI ​​agent, such as... Figure 3 As shown, it mainly includes a scheduling module, an activity selection module, a reading module, a browsing module, a following module, an observation module, and an attack module. The scheduling module sets the platform the agent belongs to, the activity time, and the waiting time. The activity selection module sets the browsing sections, section weights, and repetition counts. The browsing sections set the users to follow, topics, and keywords. The following sections set the usernames to follow. The reading section sets the reading time, and the weights of likes, favorites, comments, and shares. The agent moves according to the configured parameters.

[0064] The data storage module saves the user profile of the intelligent agent, the intelligent agent's operational behavior, the information data read, and the comment content into the database.

[0065] The corpus generation module captures user comments and popular comments under the topics of interest on the target platform, uses them as expected content, constructs a dictionary, sorts them according to weight, and selects the comments with the higher proportion as the comment content of the intelligent agent;

[0066] The behavior analysis module cleans the data in the database, uses text topic recognition algorithm, text sentiment recognition algorithm and comment opinion tendency algorithm to determine the impact of the platform's recommendation algorithm on the target intelligent agent, and visualizes it in charts;

[0067] The attribute configuration module, process customization module, data storage module, corpus generation module, and behavior analysis module are connected in sequence.

[0068] As described above, this invention can be used simultaneously on multiple platforms and with multiple intelligent agents. The description adopts a progressive approach, and similarities between different implementations can be referred to interchangeably. In some implementations, multitasking is feasible.

[0069] This invention can operate across multiple social platforms, simulate real-person behavior, run automatically on the platforms, record data, and analyze the impact of recommendation algorithms on user groups to predict user behavior trends.

[0070] This invention can simulate the behavior of normal users on major social media platforms and achieve fully automated operation. Through experimental data, the impact of recommendation algorithms on users can be analyzed, thereby obtaining ways to avoid the negative impact of recommendation algorithms on people.

[0071] The embodiments described in this specification are merely examples of implementations of the inventive concept. The scope of protection of this invention should not be considered as limited to the specific forms stated in the embodiments. The scope of protection of this invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for designing an artificial intelligence agent based on user behavior, characterized by: The method comprises the following steps: S1: configuring user static attributes for a specified artificial agent to form the identity characteristics of the agent; S2: configuring an operation process for the artificial agent to simulate the normal behavior of the user; S3: saving the behavior data of the agent during the operation of the specified agent; S4: collecting popular expectations of the platform, and sending comment content to the specified platform according to the specified parameters; S5: analyzing the influence of the recommendation algorithm on the user behavior according to the record data of the artificial agent, specifically including: S5.1: quantifying the effectiveness of the behavior of the target artificial agent by using the corrected conditional entropy and the relative index according to the behavior data of the artificial agent; S5.2: determining the behavior trend of the artificial agent by using the text topic recognition algorithm, the text sentiment recognition algorithm and the comment opinion tendency algorithm according to the record data of the artificial agent.

2. The user-behavior-based artificial intelligence agent design method of claim 1, wherein: The step S1 comprises: setting the static attributes of the artificial agent to depict the user portrait, including at least one of the following: name, gender, age, education, geographical location, interest and personal description, to form the identity characteristics of the agent.

3. The user-behavior-based artificial intelligence agent design method of claim 1, wherein: The step S2 comprises: setting the behavior parameters of the artificial agent, including daily active period, daily action frequency, weekly active days, customizing the operation mode of the agent, including scheduling module, activity selection module, reading module, recommended browsing module, collection browsing module, search browsing module, attention browsing module, attention module, observation module and attack module; and customizing agents with different behaviors according to the combination of different modules.

4. The user-behavior-based artificial intelligence agent design method of claim 1, wherein: The step S3 comprises: recording the behavior data of the agent, including action type, information data and comment content; wherein the action type includes at least one of the following: sleep, start, automatic attention / cancel attention, information reading, comment, like, collect, forward; the information data includes at least one of the following: information title, text content, text length, text type, publishing time, publisher, platform, browsing volume, comment volume, forwarding volume, like volume, collection volume and sharing volume; and the comment content includes at least one of the following: comment time, comment content, comment attitude, related text and comment user.

5. The user-behavior-based artificial intelligence agent design method of claim 1, wherein: The step S4 comprises: S4.1: grabbing the hot users and hot comment contents under the attention topic on the specified platform; S4.2: making the collected text contents into a dictionary, performing weight statistics and sorting, and generating a corpus; S4.3: according to the corresponding parameters, determining the comment content according to a random function during the comment operation, and sending the comment to the target news or video on the specified platform.

6. A system implementing the user behavior based artificial intelligence agent design method of claim 1, comprising: The attribute configuration module, the process customization module, the data saving module, the corpus generation module and the behavior analysis module; The attribute configuration module configures the user portrait of the artificial agent. Each natural person has many attributes, including name, gender, age, occupation, education, geographical location, interest and personal description. The parameter input method is used to determine the user group to be imitated by the artificial agent, thereby improving the authenticity of the artificial agent. The flow customization module selects the behavior flow of the artificial intelligence agent, including a scheduling module, an activity selection module, a reading module, a browsing module, an attention module, an observation module, and an attack module; In the scheduling module, the platform to which the agent belongs, the activity time, and the waiting time are set; In the activity selection module, the browsing board, the board weight, and the number of repetitions are set; in the browsing board, the attention user, the theme, and the keyword are set; and in the attention board, the attention user name is set; In the reading board, the reading time, the weight of the likes, the collection, the comment, and the forwarding are set; and the agent moves according to the configured parameters; The data saving module saves the user portrait of the agent, the operation behavior of the agent, the read information data, and the comment content into the database; The corpus generation module captures the user comments and the hot comments under the attention topic on the target platform, takes them as the expected content, constructs a dictionary, sorts according to the weight, and takes the comment with a high proportion as the comment content of the agent; The behavior analysis module cleans the data in the database, and uses a text theme recognition algorithm, a text emotion recognition algorithm, and a comment opinion tendency algorithm to judge the influence degree of the recommendation algorithm of the platform on the target agent; The attribute configuration module, the flow customization module, the data saving module, the corpus generation module, and the behavior analysis module are sequentially connected.

Images